Project Vision
DEDUCTION is an immersive text-based murder mystery game that combines the imaginative power of interactive fiction with the rich character development of modern narrative games. By leveraging large language models (LLMs), the project creates a deeply engaging detective experience where players must solve complex mysteries through careful investigation, character interaction, and logical deduction.
Inspiration
My inspiration for DEDUCTION came from playing Red Dead Redemption 2, where I was captivated by the depth of world-building and character-driven storytelling. What struck me most was how each character had their own motivations, backstories, and relationships that evolved throughout the game.
I wanted to bring this same level of narrative richness to text adventure games — a format that naturally stimulates the player's imagination. When you read "The crystal decanter sits on the mahogany desk, a faint smell of bitter almonds hanging in the air," each player conjures their own unique mental image, creating a personal connection to the story that visual media sometimes can't match.
The LLM Advantage
The traditional challenge with text adventures has been creating believable characters with natural dialogue. Conversations often felt scripted and limited. Large language models completely transform this limitation by enabling:
- Dynamic character interactions that respond naturally to player queries
- Contextual dialogue that evolves based on game state and previous interactions
- Psychological depth for characters with complex motivations and reactions
- Branching narratives that maintain consistency across multiple storylines
Rather than pre-writing every possible dialogue option, characters in DEDUCTION have personalities, knowledge bases, and relationship networks that inform their responses. The maid might reveal different information depending on her trust level with the player, while the family attorney might subtly alter his story if presented with contradicting evidence.
The Manor Murder
The flagship mystery, "The Manor Murder," places players in the role of an investigative journalist who awakens in Blackwood Manor to discover they're the prime suspect in a pharmaceutical magnate's murder. With a storm preventing authorities from arriving for 24 hours, players must:
- Explore the atmospheric manor and its grounds
- Interview five complex characters with their own secrets and motivations
- Collect and analyze evidence using an intuitive command system
- Recover lost memories that reveal crucial events from the night before
- Connect the pieces to identify the true killer before time runs out
This scenario features multiple investigation paths, varying character relationships based on player choices, and several possible endings determined by what evidence is discovered and which alliances are formed.
Technical Architecture
DEDUCTION employs a modular architecture with clear separation of concerns:
- Character System: LLM-powered dialogue with psychological modeling
- Evidence System: Collectable items with detailed examination options
- Command Parser: Natural language processing for player input
- State Manager: Tracks game progression and character relationships
- Memory System: Reveals flashbacks based on triggers and discoveries
- Narrative Engine: Controls story flow and ensures consistency
This structure allows for iterative development and testing, while maintaining the flexibility to enhance specific components as needed.
Learning and Innovation
Beyond creating an engaging game, DEDUCTION serves as a valuable exploration of how AI can enhance interactive entertainment. The project has provided insights into:
- LLM Prompt Engineering: Creating effective character personalities through carefully designed prompts
- Contextual AI Responses: Maintaining narrative consistency while allowing dynamic interactions
- Knowledge Management: Structuring information to create solvable yet challenging mysteries
- Narrative Design: Balancing player freedom with compelling storytelling
By working at the intersection of traditional game design and cutting-edge AI, DEDUCTION demonstrates how large language models can revitalize classic game formats for modern audiences.
Future Directions
The modular design of DEDUCTION allows for expansion beyond the initial mystery:
- Additional cases with different settings and themes
- More sophisticated character relationship systems
- Multiplayer options for collaborative investigation
- Visual enhancements like minimal illustrations or ambiance
As large language models continue to evolve, so too will the depth and realism of character interactions, creating increasingly immersive narrative experiences.
"Every murder has three elements: means, motive, and opportunity. Your job is to find all three."